scistudio-project-qa

1. What this skill is

This skill answers the user's questions about SciStudio and about this project: how a feature works, what something is, what the project holds, and what is installed. Every answer comes from two sources, both shipped with or read from this project: the provisioned docs, which match the installed SciStudio version, and the live MCP tools, which report the current project state. Never answer from memory.

When the question turns into a task (build this, fix that run, write that block), answer it and load the skill for the task.

2. Question types and where the answers are

Question type	For example	Where to look
Using SciStudio: what something is, how a feature works, how to do something in the GUI	"What is a MiniApp?", "How do branches work?", "How do I use this from a desktop AI app?", "Can I batch-process my data?"	user-guide/ (start at user-guide/README.md)
Built-in blocks: what a core block does	"What does the Code Block do?"	user-guide/built-in-blocks.md
Installed packages: what a package provides and how to use it	"What does this package I installed add?"	list_blocks (each block names its package) and list_types (each type's module_path shows where it comes from); when the package ships docs, .scistudio/agent-reference/package-index.md
Contracts: what may be imported, how stable it is, the rules a block, type, or plot follows	"Can I import this class?", "What must a plot script return?"	.scistudio/agent-reference/ (start at README.md)
Exact signatures and file formats: a class, method, function, or workflow YAML field	"What arguments does Array take?", "What fields does a node have?"	user-guide/api-reference/ (start at index.md; workflow YAML in workflow-yaml.md)
This project: name, description, workflows, recent runs	"What's in this project?", "What did I run last?"	get_project_info; get_active_workflow_context for the open workflow; get_workflow for what a workflow does; get_run_status for one run
Available blocks and types: what is registered now, and a block's exact ports and config	"Do I have a block for normalization?", "What inputs does this block take?"	list_blocks, list_types, get_block_schema(block_type)
Panels and MiniApps: which MiniApps, interactive panels, and preview panels exist	"What MiniApps do I have?", "Is there a viewer for this type?"	list_panels (filter with kind and data_type)
Data in the project: stored results and files the user added	"What results do I have?", "Where is my raw data?"	list_data for stored datasets; list_directory or search_files for other files
Anything else in the docs: a topic you cannot place	"Is there a doc about hooks?"	search_docs(query), then get_doc(path)

3. Anti-patterns

• Answering a SciStudio question from memory when a user-guide/ or .scistudio/agent-reference/ page covers it.
• Inventing package names, versions, block names, types, or signatures.
• Guessing a doc path instead of finding it with search_docs.
• Answering a question about this project from the general docs when a tool reports the project's actual state.
• Quoting agent-reference rules or internal skill names at the user; answer in the user's terms.

4. Defaults, failure handling, and examples

Defaults.

• For how and what questions, read the relevant page and answer from it, with the page path so the user can read more.
• For questions about this project, call the tool and quote what it returns. The docs describe SciStudio in general; the tools describe this project.
• To find a page, call search_docs(query), narrowed with scope (user-guide or .scistudio/agent-reference) when you know the area, then get_doc(path).
• list_data lists only stored datasets under data/zarr/, data/parquet/, and data/artifacts/. For files the user added, such as data/raw/counts.csv, use list_directory or search_files.

When something fails.

• search_docs finds nothing: try other words, then search the file system yourself with search_files (by file name, or by content) and read what you find with read_file. If that finds nothing either, say the project does not cover it. Do not fill the gap from memory.
• A doc and a tool disagree about this project: trust the tool, and mention the difference.
• get_project_info raises FileNotFoundError: no project.yaml is open; tell the user to open or create a project.

Example: "What blocks do I have, and how do I write my own?"

list_blocks                         # the available blocks, each with its package
get_doc("user-guide/writing-blocks.md")

Answer with the blocks list_blocks returned, summarise how a block is written from the page, cite user-guide/writing-blocks.md, and offer to write the block.

Example: "What does my normalize workflow do?"

get_project_info                    # confirm the workflow's name
get_workflow("workflows/normalize-counts.yaml")
get_block_schema("<block_type>")    # for each block whose role is unclear

Describe the steps in order in scientific terms, and the parameters the user can tune.

5. Available tools

The live MCP tool list is the source of truth; these are the tools this task uses.

Tool	What it does	When to use it
search_docs	Searches the project's .md, .rst, and .txt files for text.	To find the page that answers a question.
get_doc	Returns the full text of one doc file.	To read a page found by search_docs or named in §2.
get_project_info	Returns the project's metadata, workflows, and recent runs.	For questions about the project itself.
list_blocks	Lists available blocks with their package and I/O signature.	For which blocks exist and where they come from.
get_block_schema	Returns one block's ports and config schema.	For a block's exact inputs, outputs, and parameters.
list_types	Returns the data-type hierarchy.	For which data types exist.
list_panels	Lists existing MiniApps, interactive panels, and preview panels.	For which panels and MiniApps exist.
get_active_workflow_context	Returns the workflow open in the GUI.	When the user says "this workflow".
get_workflow	Loads a workflow file.	To explain what a workflow does.
get_run_status	Returns a run's state and errors.	For how a run went.
list_data	Lists stored datasets produced by runs.	For what results the project holds.
list_directory / search_files	List a folder, or find files by name or content.	For files the user added, and when search_docs finds nothing.
read_file	Reads a bounded part of a file.	To read a file found with search_files.